Abstract
There are limited research contributions targeting sentiment analysis in feedback in Arabic gulf dialect, in particular, and the Arabic language in general. Furthermore, the inadequate and limited adoption of classification techniques and natural language processing is noticeable in the sentiment analysis projects addressing the Arabic language. Hence, this paper focuses on analyzing the sentiments in automobile and real estate domains through the application of the state-of-the-art word-embedding model "BERT" and a collection of deep learning models (GRU, LSTM, CNN, CNN-GRU and BiLSTM). The results of classification revealed that combining the BERT with deep learning models have shown efficiency in analyzing sentiments and yielded outstanding results.